Question

Using TMM normalized counts in Deseq2

0

Entering edit mode

mmicky5050 • 0

@829aeee2

Last seen 3.8 years ago

Hi I am using Partek to analyze RNA-seq data, Partek has a median normalization step before Deseq2, which works fine. I understand Deseq2 takes raw data, but Partek has separated the normalization step recently (previously it normalization used to be part of Deseq2 in Partek).

While playing with the various normalizations, I have noticed that if I use TMM normalization followed by Deseq2, I get much better results. I know TMM is for EdgeR but unfortunately Partek does not have EdgeR.

Is there anyway, one can get away with using TMM normalized input for a modified Deseq2, which has its median normalization disabled?

Thanks

```

RNAseq123 • 5.3k views

ADD COMMENT • link written 3.9 years ago by mmicky5050 • 0

score 0 · Answer 1 · 2021-01-09

0

Entering edit mode

mmicky5050 • 0

@829aeee2

Last seen 3.8 years ago

ok i found the answer.

This paper shows that TMM normalization (which is the default normalization method associated with EdgeR) is not suitable for use with DESeq2:

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0157022

They go into detail in their supplementary:

https://doi.org/10.1371/journal.pone.0157022.s001

ADD COMMENT • link 3.9 years ago mmicky5050 • 0

1

Entering edit mode

Actually TMM is in principle perfectly suitable to be used with DESeq2. Any of the scale normalization methods can be used with any of the packages. You simply have to convert norm.factors from edgeR into size factors for DESeq2, as the paper you have linked to points out. You cannot simply treat norm factors and size factors as equivalent because they are on different scales. It is just a programming detail rather than any issue of incompatibility.

ADD REPLY • link 3.8 years ago Gordon Smyth 51k